In [79]:
# Importing necessary libraries , rest of the libraries imported as and when needed
import numpy as np
import pandas as pd
%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns
from scipy import stats
from scipy.stats import zscore
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score,confusion_matrix
In [2]:
df_vehicle=pd.read_csv('vehicle.csv') # reading data file
In [3]:
df_vehicle.head() # viewing first 5 records
Out[3]:
In [4]:
df_vehicle.describe() # exploring the statistical features
Out[4]:
In [5]:
df_vehicle.shape
Out[5]:
In [6]:
df_vehicle.columns.T # Seeing column name in data frame
Out[6]:
In [7]:
df_vehicle.info()
In [8]:
df_vehicle.isna().sum() # looking for missing values , so those can be imputed
Out[8]:
In [9]:
#df_vehicle.isnull().sum() # This is same as 'na' values
In [10]:
df_vehicle[df_vehicle['circularity'].isnull()==True]
Out[10]:
In [11]:
df_vehicle['class'].unique()
Out[11]:
In [12]:
df_vehicle.groupby('class').count() # checking number of values for each class
Out[12]:
In [13]:
df_vehicle.groupby('class').median() # seeing median for each feature respective to the class
Out[13]:
In [57]:
df_vehicle.groupby('class').mean() # seeing mean for each feature respective to the class
Out[57]:
In [15]:
for col in df_vehicle.columns:
if col == 'class':
break
df_vehicle[col]=df_vehicle.groupby('class')[col].transform(lambda x: x.fillna(x.median()))
In [16]:
#df_vehicle['circularity']=df_vehicle.groupby('class').circularity.transform(lambda x: x.fillna(x.median()))
In [17]:
df_vehicle.replace({'class': {'bus' : 0,'car' : 1,'van' : 2}},inplace=True) # assiging numeric value to target class so object can be converted to numeric
df_vehicle['class']=pd.to_numeric(df_vehicle['class'])
df_vehicle.head(10)
Out[17]:
In [18]:
df_vehicle.boxplot(figsize=(30,20),); # looking for boxplot,to determine outliers in various features.
# It is evident that there are high number of outliers in certain features
In [19]:
sns.boxplot(y='scaled_variance.1',x='class',data=df_vehicle)
Out[19]:
In [20]:
sns.boxplot(y='scaled_variance',x='class',data=df_vehicle)
Out[20]:
In [21]:
sns.boxplot(y='radius_ratio',x='class',data=df_vehicle)
Out[21]:
In [22]:
sns.scatterplot(x='radius_ratio',y='class',data=df_vehicle)
Out[22]:
In [23]:
for col in df_vehicle.columns:
if col == 'class':
continue
quart75=np.percentile(df_vehicle[col],75)
quart25=np.percentile(df_vehicle[col],25)
iqr=quart75-quart25
l_point=quart25-(1.5*iqr)
u_point=quart75+(1.5*iqr)
df_vehicle.loc[(df_vehicle[col] < l_point) | (df_vehicle[col] > u_point),col]=df_vehicle[col].median()
In [24]:
df_vehicle.boxplot(figsize=(50,20),); # Below boxplot show there are no outliers in the data now
In [25]:
#df_vehicle[df_vehicle['circularity'].isnull()==True]
In [26]:
#df_vehicle.iloc[[5,105,118],]
In [27]:
import seaborn as sns
sns.pairplot(df_vehicle,hue='class')
Out[27]:
In [28]:
df_vehicle.info()
In [29]:
X=df_vehicle.drop('class',axis=1) # Assinging all features to X except the target variable, hence demarcating
Y=df_vehicle[['class']] # Assiging only target feature to Y
In [30]:
Y.shape
Out[30]:
In [31]:
X_std=X.apply(zscore) # Standardizing the data using zscore
In [32]:
plt.figure(figsize=(15,10))
sns.heatmap(X_std.corr(),annot=True) #Plotting the correlation matrix depicting features relationship
Out[32]:
In [33]:
X_train_std,X_test_std,Y_train,Y_test=train_test_split(X_std,Y,test_size=0.25)
In [34]:
cov_matrix = np.cov(X_train_std.T) #Finidng covariance matrix
print('Covariance Matrix \n%s', cov_matrix)
In [35]:
eig_vals, eig_vecs = np.linalg.eig(cov_matrix) #Finiding eigen value and vector. There will be 14 eigen values and 14 list of eigen values
In [36]:
print('Eigen Vectors \n%s', eig_vecs)
print('\n Eigen Values \n%s', eig_vals)
In [37]:
eigen_pairs = [(np.abs(eig_vals[i]), eig_vecs[ :, i]) for i in range(len(eig_vals))] #Creating eigen pairs, at 0 index it will be eigen value and at index 1 it will be eigen vector
print(eigen_pairs)
In [38]:
eigen_pairs.sort(key=lambda x: x[0], reverse=True)
In [39]:
# Visually confirm that the list is correctly sorted by decreasing eigenvalues
print('Eigenvalues in descending order:')
for i in eigen_pairs:
print(i[0])
eigen_pairs
Out[39]:
In [40]:
tot = sum(eig_vals)
var_exp = [( i /tot ) * 100 for i in sorted(eig_vals, reverse=True)]
cum_var_exp = np.cumsum(var_exp)
print("Cumulative Variance Explained", cum_var_exp)
In [41]:
plt.figure(figsize=(10 , 7))
plt.bar(range(18), var_exp, alpha = 0.5, align = 'center', label = 'Individual explained variance')
plt.step(range(18), cum_var_exp, where='mid', label = 'Cumulative explained variance')
plt.ylabel('Explained Variance Ratio')
plt.xlabel('Principal Components')
plt.legend(loc = 'best')
plt.tight_layout()
plt.show()
In [44]:
eigvalues_sort = [eigen_pairs[index][0] for index in range(len(eig_vals))]
eigvectors_sort = [eigen_pairs[index][1] for index in range(len(eig_vals))]
In [45]:
P_reduce = np.array(eigvectors_sort[0:12]).transpose() # Selecting first 12 eigen vectors of all 18
Proj_train_data = np.dot(X_train_std,P_reduce) # projecting training data onto the eight eigen vectors
Proj_test_data = np.dot(X_test_std,P_reduce) # projecting test data onto the eight eigen vectors
In [47]:
# complete matrix operation
print(Y_train.shape)
X_train_std.shape, P_reduce.shape, Proj_train_data.shape
Out[47]:
SVM¶
In [69]:
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV
In [70]:
svc_model=SVC()
parameters={'kernel' : ('rbf','linear'),
'C' : [0.01, 0.05, 0.5, 1]}
In [71]:
clf = GridSearchCV(svc_model, parameters)
In [72]:
clf.fit(X_train_std,Y_train)
Out[72]:
In [73]:
print("best hyperparameter from the values iserted {}".format(clf.best_estimator_))
In [74]:
sorted(clf.cv_results_.keys())
Out[74]:
In [75]:
print ("Best score on training set {}".format(clf.best_score_))
In [78]:
prediction=clf.predict(X_test_std) #Prediction on X_test
In [80]:
print("printing confusion matrix")
confusion_matrix(Y_test,prediction)
Out[80]: